Conditional Viability of Refurbished EV/PHEV Batteries: A Risk-Informed Decision Framework for Circular Pathway Selection

Abstract

End-of-life electric-vehicle and plug-in hybrid (EV/PHEV) battery packs pose a recurrent decision: refurbish, redeploy in second-life storage, recycle, or reject. Technical condition, safety, economics, regulation, traceability, and environmental benefit interact, making pathway selection a systems-level decision problem. This paper develops a risk-informed multi-criteria framework for the conditional viability of refurbished batteries under data-scarce conditions. Failure mode, effects, and criticality analysis (FMECA) supplies a pathway-specific residual-risk penalty; multi-criteria decision analysis (weighted-sum and the Technique for Order of Preference by Similarity to Ideal Solution, TOPSIS) orders four alternatives on six benefit criteria; and a screening-level avoided-burden indicator, not a life-cycle assessment, positions the environmental criterion. An Integrated Viability Index (IVI) offsets weighted benefits against the risk penalty through one tunable coefficient. All inputs are illustrative and literature-informed; the demonstration tests decision logic, not empirical pathway performance. Preference is conditional: refurbishment leads under economic and technical priority with credible risk mitigation, second-life reuse under environmental priority, and recycling under safety, regulatory, and infrastructure constraints, while rejection never leads. As the risk penalty rises, leadership migrates traceably toward recycling, and IVI–TOPSIS divergence localizes exactly where the risk treatment changes the decision. A proposed Refurbished-Battery Suitability Index (RBSI) would couple measured diagnostics to the IVI; its calibration remains future work.

1. Introduction

Every retired traction battery forces a choice that current practice makes largely ad-hoc [1]: restore the pack, redeploy it in a less demanding role, recover its materials, or scrap it. Packs sold during the first wave of electric-vehicle (EV) and plug-in hybrid electric-vehicle (PHEV) adoption are now approaching the end of their first automotive service life [2,3], and how they are handled is central to the circular-economy ambition of retaining embodied value and energy rather than discarding them [4]. Value-retention strategies form an ordered hierarchy, from refurbishment and remanufacturing through second-life redeployment to material recycling [4,5,6,7], and each pathway carries distinct economic, environmental, and safety implications [1,8]. The reported benefits are conditional rather than intrinsic: second-life use can lower levelized storage cost and life-cycle greenhouse-gas emissions relative to new systems under favorable conditions [9,10]; remanufacturing can deliver environmental advantages in specific settings, set against the substantial manufacturing carbon intensity of lithium-ion cells [11,12,13,14,15]; consumer acceptance of refurbished products is governed by trust instruments under information asymmetry [16,17,18,19]; and the regulatory architecture around Regulation (EU) 2023/1542 and the Digital Battery Passport (DPP) is still taking shape [20,21,22,23]. Two operational obstacles sharpen the decision. First, residual capacity is routinely treated as a sufficient proxy for battery health, with the commonly cited ~80% state-of-health (SoH) retirement convention used as a single pass/fail criterion [24]; SoH is in fact multidimensional—impedance growth, thermal behavior, inter-cell homogeneity, round-trip efficiency, and self-discharge govern usable power, safety margin, and the realizable second-life trajectory [25,26,27]—so capacity-only screening can admit unsafe packs and exclude suitable ones [24], and a defensible decision requires a multimodal diagnostic basis coupled to an auditable decision rule. Second, feasibility is set by the surrounding system as much as by the pack: end-of-life detection, aggregation, and the regulated transport of used lithium-ion batteries constrain every pathway upstream of any choice [1,8]. Confronted with a specific used pack, the question facing a fleet operator, dismantler, refurbisher, or policymaker is therefore decisional: when should the pack be refurbished, redeployed in a second-life application, recycled, or rejected for safe disposal? Answering it requires the simultaneous weighing of technical residual performance, safety, economic viability, environmental benefit, regulatory readiness, and market or implementation readiness, together with the operational and safety risks that attach to each pathway.

The building blocks of such a decision are individually well studied, but they have developed largely in isolation. Degradation research characterizes second-life potential and its conditionality on residual condition and host application [2,9,10]; life-cycle studies quantify manufacturing burdens and the settings in which remanufacturing pays off environmentally [11,12,13,14,15]; recycling research documents process challenges and dedicated recovery technologies [3,6,7]; consumer research examines trust and acceptance in refurbished-product markets [16,17,18,19,28]; and reverse-supply-chain and trade-in models formalize the flows and the cost and demand conditions under which life extension is economically preferable [5,29]. Where pack-level data are scarce, structured methods substitute transparent reasoning for deterministic estimates: failure mode, effects, and criticality analysis (FMECA) prioritizes risk [30,31], and multi-criteria decision analysis (MCDA) methods—the weighted sum, the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) [32], VIKOR [33], the analytic hierarchy process (AHP) [34], and the best-worst method (BWM) [35]—rank alternatives under conflicting criteria, with design science framing artifact construction [36] and Delphi elicitation the disciplined route to calibration under uncertainty [37]. These strands rarely meet in a single decision rule. FMEA–MCDA hybrids apply MCDA operators to rank failure modes rather than pathways [31]; circular automotive decision studies map decision contexts without a reproducible composite index [1]; MCDA pathway treatments rank alternatives without a separately computed failure-risk penalty [32,33]; and a recent lifecycle-management framework allocates retired packs on technical indicators and passport-based traceability without stakeholder-weighted benefit criteria or an explicit residual-risk penalty [38]. The closest recent contribution, by Ma et al. [39], optimizes reuse-versus-recycling allocation of retired lithium-ion batteries on measured economic and environmental functions, with chemistry-specific (lithium iron phosphate (LFP) and nickel–manganese–cobalt (NMC)) and state-of-health-differentiated data and a service-based functional unit; it is empirically grounded where the present framework is deliberately screening-level, but it does not model safety, regulation, or traceability as decision criteria, carries no explicit pathway-level residual-risk penalty, and does not treat automotive-grade refurbishment or rejection as distinct alternatives. Within the corpus assembled by the scoping synthesis of Section 2.2, we did not identify a framework that jointly combines technical risk, safety, economic viability, environmental impact, refurbishment feasibility, traceability, regulation, infrastructure, and market acceptance to choose among refurbishing, reusing, recycling, or rejecting an EV/PHEV battery within one transparent, reproducible decision logic;

Table 1. Positioning of the present framework against the closest integrative contributions in the literature. Symbols: ● fully addressed, ◗ partially addressed, ❍ not addressed. The assessment is qualitative, refers to each work’s primary scope, and is asymmetric by construction: the cited works pursue empirical or optimization depth within narrower boundaries, whereas the present framework pursues breadth at screening level; “fully addressed” therefore means explicitly represented in the decision architecture, not empirically resolved. The lifecycle-management framework of [38] is discussed in the text rather than tabulated.

Capability	Reverse-Supply-Chain/Trade-In Models [5,29]	Circular Automotive Decision Studies [1]	MCDA Pathway Ranking [32,33]	FMECA Risk Studies [30,31]	Pathway-Allocation Optimization [39]	This Paper
Technical/residual-performance criterion	◗	◗	●	❍	◗	●
Inherent/designed safety as explicit criterion	❍	◗	◗	◗	❍	●
Economic viability	●	◗	◗	❍	●	●
Environmental benefit criterion	❍	❍	◗	❍	●	●
Regulation/traceability (DPP)	❍	◗	❍	❍	❍	●
Market/implementation readiness	◗	●	◗	❍	❍	●
Separately computed failure-risk penalty	❍	❍	❍	●	❍	●
Choice across all four pathways	◗	◗	◗	❍	◗	●
Reproducible composite index	❍	❍	◗	❍	❍	●

positions the present framework against the closest integrative contributions. The gap addressed here is therefore not the absence of studies on individual pathways or methods [1,5,29,31,32,33,38,39], but the lack of an integrated, risk-penalized decision architecture that links battery-level diagnostic eligibility to pathway-level viability under data scarcity—most acutely in emerging-market settings, where the regulatory enablers of trust, such as the EU Digital Product Passport [21], and the readiness of the local market to act on them may be temporally mismatched (a motivating assumption delimited in Section 2.6, not an empirical claim about any specific market).

This paper develops a risk-informed multi-criteria decision-support framework for assessing the conditional viability of refurbished EV/PHEV batteries within circular-economy systems: FMECA-based risk prioritization [30,31] characterizes each pathway’s failure exposure, MCDA ranking (weighted-sum and TOPSIS) [32,33] orders the four alternatives on six benefit criteria, and a screening-level avoided-burden proxy anchored in published life-cycle data [15,40,41] positions the environmental criterion. Five research questions structure the development: RQ1, which technical, safety, economic, environmental, and regulatory criteria determine the conditional viability of refurbished EV/PHEV batteries; RQ2, how FMECA can identify and prioritize the failure modes affecting the refurbishment, second-life, and recycling pathways; RQ3, how MCDA methods can rank circular battery pathways under conflicting criteria; RQ4, how a screening-level indicator anchored in published life-cycle data can be embedded without claiming a full ISO-compliant life-cycle assessment; and RQ5, what measurement and elicitation architecture would render the framework empirically testable. The contribution is integrative rather than method-level the components are established and used as published, and the novelty lies in their coupling, sequencing, and decision architecture. Five design elements carry it: benefits and risk enter the final index as distinct terms, six stakeholder-weighted benefit criteria forming a weighted score from which a separately computed FMECA residual-risk penalty is subtracted in the Integrated Viability Index (IVI, Equation (15); the risk-penalty coefficient $\lambda$ is explicit and tunable, so the dependence of pathway preference on the credibility of risk-mitigation instruments (warranty, independent SoH certification, battery-passport traceability [20,21,23]) is an inspectable parameter rather than an implicit assumption; a battery-level Refurbished-Battery Suitability Index (RBSI) is proposed—specified, not executed—to connect multimodal diagnostics to the pathway-level IVI, composite indexing itself not being claimed as novel [42]; data scarcity is met with declared, illustrative input matrices released together with the scripts and sensitivity analyses, so every reported result can be regenerated and its dependence on the assumptions traced; and the comparison spans all four end-of-life pathways, including rejection, so the framework can conclude that no value-retaining pathway is defensible for a given pack. The calibration protocol (Delphi-based weighting, empirical diagnostics, and an ISO-compliant life-cycle assessment) is specified as future work. The paper proceeds as follows: Section 2 sets out the design-science methodology and the conceptual framework, separating completed work from deferred calibration protocols; Section 3 formalizes the model as Equations (2)–(16), the battery-level RBSI having been defined as Equation (1); Section 4 demonstrates the model on illustrative inputs; Section 5 discusses theoretical, practical, and policy implications; Section 6 states the limitations and the validation agenda; and Section 7 concludes.

2. Methodology and Framework

This section sets out the research design and develops the conceptual framework; the calibration protocols deferred to future work are summarized here and specified in full in the Supplementary Materials.

2.1. Research Design and Rationale

The study follows a design-science research paradigm [36]: its output is a decision-support artifact—a risk-informed multi-criteria framework for the conditional viability of refurbished EV/PHEV batteries—not a hypothesis test on primary data. The work proceeds in three stages. Stage 1 derives the decision alternatives, criteria, and failure modes from a structured scoping synthesis (Section 2.2). Stage 2 formalizes the FMECA–MCDA–screening model and its Integrated Viability Index (Section 2.4, Section 2.5, Section 2.6, Section 2.7 and Section 3) and characterizes its behavior through scenario and sensitivity analysis (Section 3.9 and Section 4). Stage 3, expert calibration through a modified Delphi procedure with AHP/BWM weight elicitation, is specified as a protocol but not conducted (Section 2.3; full instrument in the Supplementary Materials), so all numeric inputs remain illustrative and literature-informed. The sequencing is deliberate, not a concession: where field data on refurbished packs are scarce, a transparent structure with explicitly assumed inputs is easier to scrutinize, and therefore more defensible, than point estimates whose precision the evidence cannot support [43,44,45].

2.2. Stage 1: Structured Literature Synthesis

The decision elements were derived from a structured, scoping-style synthesis of peer-reviewed work on EV/PHEV battery degradation and second-life potential; refurbishment, reuse, and recycling pathways; circular-economy value retention; battery safety and risk; traceability and regulation; and multi-criteria decision methods (synthesized in Section 1). A scoping logic was chosen over a full systematic review because the purpose was to map recurring decision elements, not to estimate an effect size, and because the evidence base is fragmented across the engineering, environmental-assessment, operations-research, and policy literatures.

Sources were drawn from Scopus, Web of Science, and Google Scholar and complemented by standards (ISO, IEC, UL, UN), regulatory texts (EUR-Lex), and laboratory grey literature (e.g., NREL, PNNL). The search window was 2018–2026, with seminal earlier methodological works retained; inclusion was limited to peer-reviewed studies, standards, or agency reports addressing end-of-life EV/PHEV battery pathways and their technical risk, economics, environmental impact, or decision methods (search strings and exclusion criteria in the Supplementary Materials). From the retained corpus, the four decision alternatives (A1–A4; Section 2.5 and Section 3.1) and the six evaluation criteria (C1–C6; Section 2.4 and Section 3.2) emerged as the recurring options and decision dimensions, and the candidate failure modes as the recurring hazards. One limitation must be stated plainly: screening counts and a flow diagram are not reported, so the synthesis is a documented scoping exercise, not a PRISMA-compliant systematic review, and the omission of relevant frameworks cannot be ruled out (Section 6).

2.3. Proposed Expert Calibration and Diagnostic Sublayer (Future Work)

Two future-work tracks would replace the illustrative inputs with elicited and measured values; both are specified in full in the Supplementary Materials, and neither has been implemented. The first track is a modified Delphi procedure [37]. A purposive, heterogeneous panel of roughly 12 to 20 experts would work through three to four anonymous rounds, first refining the criteria and failure-mode inventory and then eliciting three classes of judgment: criterion weights, through Analytic Hierarchy Process (AHP) pairwise comparisons or the Best-Worst Method (BWM) [34,35]; severity, occurrence, and detection (S/O/D) ratings on the 1–10 scale; and plausible ranges for the coefficient $\lambda$ (Equation (15)) and for the SoH thresholds. To preserve the C2-versus-risk partition in elicitation, the protocol would score C2 and S/O/D in separate blocks with different anchors: C2 captures designed safety and control maturity, while severity captures consequence conditional on failure despite designed controls; consistency checks would then flag responses where high severity appears to depress C2 mechanically. Consensus and stability rules would be fixed a priori. Any implementation would require informed consent, anonymization, and institutional ethics approval; as no expert data have been collected, the Institutional Review Board and Informed Consent statements are recorded as “Not applicable.”

The second track is a battery/cell-level diagnostic sublayer: four sequential layers of increasing depth. Layer 1 applies rapid non-destructive screening, with exclusion criteria that remove unsafe or untraceable units before further testing. Layer 2 adds confirmatory electrochemical and thermal diagnostics (reference capacity, pulse-power and internal-resistance characterization, impedance spectroscopy, thermal response under load, self-discharge, inter-cell homogeneity, and round-trip efficiency), with no single indicator treated as sufficient [25,26,27]. Mechanistic attribution—for example, loss of lithium inventory, loss of active material, or impedance growth consistent with plating—is inferred from such measurements rather than observed directly, and the protocol keeps observed indicators distinct from inferred mechanisms [27,46]. Layer 3 performs active refurbishing and reconfiguration, in a comparative design against passive regrouping and no refurbishing [47]. Layer 4 estimates the second-life trajectory and remaining useful life, with the associated uncertainty [46,48]. The layers are designed to align with UL 1974 [49] and the related safety, EV-test, and transport standards [50,51,52,53,54,55,56]. Their output is eight standardized, traceable estimates—each carrying unit identity at cell, module, or pack level, test conditions, instrument calibration status, and timestamps, so that records remain compatible with Digital Battery Passport documentation [20,21] and open battery-data formats [57]—one per RBSI component of Section 2.7; these would supply the measured inputs for the battery-level index, whose components and failure characterization in turn feed the pathway-level IVI. The accompanying statistical-validation plan, the proposed ISO 14040/14044-compliant life-cycle assessment [40,41] and the techno-economic and risk assessment [40,41,57,58,59,60,61] are likewise specified in the Supplementary Materials.

2.4. Conditional-Viability Logic

The framework rests on a single premise: the viability of a refurbished EV/PHEV battery is not an intrinsic property of the pack. It is a conditional outcome, determined jointly by the residual technical and safety state, the relative weighting of the decision criteria, and the credibility of the available risk-mitigation mechanisms. The question is therefore not whether refurbishment is viable in the abstract, but under which conditions each pathway becomes the preferred allocation of a specific pack. Framed this way, the problem ceases to be a technological assessment of state of health and becomes a decision problem in which technical condition is one input among several conflicting criteria [1].

Conditional viability has three coupled dimensions. The first is benefit: six criteria (C1 technical viability; C2 inherent or designed safety and reliability; C3 economic viability; C4 environmental benefit; C5 regulatory readiness; C6 market or implementation readiness), each scored so that higher is better and reconciled through MCDA [32,33]. Risk is the second dimension. FMECA [30,31] identifies the residual operational and safety failure exposure specific to each pathway; this exposure enters as a separate penalty, not a seventh weighted criterion, so a pathway that scores well on benefits can still be demoted for disproportionate failure exposure. Mitigation, the third dimension, conditions the interpretation of the second: how much weight operational risk deserves depends on the instruments available to contain it—warranty schemes, independent SoH certification, and Digital Product Passport (DPP) traceability [20,21,22,23]. The strength of the risk penalty is an interpretable parameter, not a fixed constant.

A natural objection is double counting: safety appears on both sides of the architecture. The two are distinct constructs: the benefit criterion C2 rewards a pathway’s inherent or designed safety, whereas the FMECA penalty captures the residual operational exposure that survives designed controls. The partition is formalized in Section Construct Partition: Inherent Safety (C2) Versus Residual Operational Risk ($R_i$) and tested in Section 4.3.

Integrating benefit and risk yields the conceptual core of the framework: a viability score that rewards multi-criteria performance while penalizing pathway-specific failure exposure. Section 3 formalizes this score as the IVI in Equation (15); the risk-penalty coefficient $\lambda$ governs the trade-off and is where the mitigation dimension enters the formal model. The implication examined in Section 4 follows directly: no pathway dominates universally, because the preferred alternative shifts with the criterion weighting and how strongly residual risk is penalized.

One boundary of the illustrative implementation must be made explicit. The performance matrix exercised in Section 3 and Section 4 is pathway-generic: its scores characterize the pathways, not an individual pack, so pack-level discrimination currently enters only through the threshold gates of Equation (16). Pack-specific viability assessment becomes possible once the RBSI coupling of Section 2.7 supplies measured, unit-level inputs; until then, the framework ranks pathways for a class of packs rather than a particular unit.

2.5. Pathway Hierarchy and Decision Nodes

The four end-of-use options are arranged as a hierarchy of value-retention pathways, ordered by the embodied value and manufacturing energy each preserves [4]. Refurbishment (A1) occupies the top: diagnostics, cell or module replacement, and reconditioning restore the pack for continued automotive-grade reuse (refurbishment restores a specified functional condition, whereas remanufacturing returns the pack to as-new specification [4,11]). The position is double-edged. A1 retains the largest share of embodied material and manufacturing energy and, to the extent that the restored pack actually displaces new-pack production, avoids part of the associated manufacturing carbon intensity [11,12]; yet it also concentrates the highest operational and safety exposure because the pack must again meet demanding automotive duty cycles under the information asymmetry discussed in Section 1 of this manuscript and supported by [16]. Second-life reuse (A2) resolves the tension differently: redeploying the degraded pack in a less demanding stationary application extends its service life and may lower levelized storage cost and life-cycle emissions relative to new systems, by application-dependent margins [9,10]. Recycling (A3) recovers materials when reuse is infeasible or unsafe: the lowest tier of value retention, but the most controllable and regulation-aligned end state, with the least operational risk, though recovery yields and process impacts vary with feedstock and method [3,6,7]. Rejection or safe disposal (A4) destroys the remaining value: invoked only when residual risk precludes every value-retaining pathway, it serves as the lower bound for comparing the three productive pathways.

This ordering is a default, not a verdict. Reverse logistics, end-of-life detection, and regulated transport condition the practical accessibility of all four pathways [1,8]: end-of-life lithium-ion packs move as regulated dangerous goods subject to packaging and testing requirements [8,56], custody and provenance must remain documented across handoffs [20,23], and feedstock aggregation, transport distance, and the location and capacity of diagnostic, refurbishment, and recycling infrastructure set the practical envelope within which any pathway operates [1,8]. The framework therefore does not apply the hierarchy mechanically: it is the structure over which the conditional-viability logic operates, and realized preference may depart from the default whenever criterion weights or risk penalties justify the departure.

The conditional-viability logic itself is enforced through decision nodes that screen each pack before any composite scoring (formalized in Section 3.8). A safety and risk gate comes first: a pack whose failure exposure is excessive or whose safety state is critically low is routed directly to rejection (A4). The gate is deliberately non-compensatory—no economic or environmental benefit can justify retaining a pack whose residual risk cannot be credibly contained. A pack that clears the gate reaches a technical-condition node governed by residual SoH, which directs it toward refurbishment (A1), second-life reuse (A2), or recycling (A3) as SoH falls through two thresholds. Stated as eligibility conditions: refurbishment is admissible only after the safety and traceability gates are cleared and restoration of function can be verified; second-life reuse applies when automotive requirements are no longer met but a less demanding application remains safe; recycling applies when restoration or reuse is unsafe or infeasible and material recovery is the justified route; and rejection remains the non-negotiable safety backstop. The division of labor is by design: the nodes provide a transparent, rule-based first pass, and the IVI ranks the survivors, which are still scored on all six benefit criteria and penalized by their FMECA-derived risk. The threshold values are illustrative parameters, to be calibrated against field data before operational use.

2.6. Emerging-Market Constraints and Infrastructure Readiness

The framework is intended for data-scarce, emerging-market contexts; its design assumes that conditions a mature market takes for granted—certified diagnostics, established reverse logistics, enforceable traceability—cannot be presumed there. Three constraints follow and shape its intended use.

The most tangible constraint is infrastructure. Pathway accessibility depends on physical and logistical capacity: diagnostic facilities for SoH estimation, refurbishment workshops, deployment channels for stationary storage, and compliant recycling and transport infrastructure [8]. Where this capacity is thin, the upper value-retention pathways may be effectively unavailable even for technically suitable packs [1]. The response is twofold: market or implementation readiness enters as an explicit criterion, and an infrastructure-constrained weighting is examined as one analytical scenario.

Regulatory and traceability readiness poses a different difficulty. The carbon-footprint declaration and Digital Battery Passport obligations of Regulation (EU) 2023/1542—the passport obligation applying from February 2027 to EV and light-means-of-transport batteries and to industrial batteries above 2 kWh—together with blockchain-based certification [20,21,22,23], are the instruments on which trust in refurbished packs would rest; in emerging-market settings, however, they may be mandated before the market and infrastructure capable of acting on them exist. This possible mismatch is a contextual assumption motivating the design, not an empirical claim about any specific market ([21] is an EU-level analysis). The framework accommodates the mismatch by treating regulatory readiness as a distinct criterion and letting the risk-penalty strength reflect the practical, rather than nominal, credibility of trust mechanisms.

The third constraint is behavioral: information asymmetry and trust. Where condition and usage history are not independently certified, perceived condition dominates purchase decisions [16,19]. The evidence, drawn largely from refurbished consumer electronics, indicates a trust hierarchy in which reputation, price, warranty, and quality assurance are decisive, with eco-certification mattering mainly to environmentally oriented buyers [16,17,18,19]. The framework transfers this pattern to EV/PHEV batteries by analogy, not measurement: the corresponding instruments—warranty, independent SoH certification, DPP-based transparency, and service quality—are the risk-mitigation mechanisms of Section 2.4, and their practical effectiveness modulates the risk penalty. Where they are weak, the design implies a stronger penalty, shifting the conditional preference away from refurbishment toward more controllable end states such as recycling.

None of these constraints can be engineered away by a decision framework; what it can do is keep its assumptions explicit and adjustable, hence the reliance on structured decision methods [30,31,32,33,34,35,36,43,44,45]. One element remains: the composite index that couples the proposed measurement sublayer (Section 2.3) to the IVI.

2.7. The Refurbished-Battery Suitability Index and Its Coupling to the IVI

The four-layer protocol of Section 2.3 feeds a single pipeline that connects measurement to strategic decision: multimodal diagnostics → RBSI → IVI → techno-economic, life-cycle, and risk assessment. The layers would produce standardized, traceable estimates of the eight health-and-eligibility dimensions, to be aggregated at the cell or pack level into the proposed Refurbished-Battery Suitability Index (RBSI); the same dimensions would supply measured values for the technically measurable elements of the IVI benefit criteria C1–C6, while the failure-mode characterization informs the penalty term $R_i$ (Equation (15)). Battery-level eligibility would thereby be converted into pathway-level viability, with uncertainty to be propagated from the technical layer through to the economic and environmental outputs. The point of the coupling is empirical testability: once the sublayer is implemented, measured values of known provenance could replace the IVI’s presently illustrative inputs.

The RBSI is proposed as a battery- or cell-level composite eligibility metric that synthesizes multimodal diagnostic evidence into a single, interpretable decision-support score. For unit i,

$${\mathrm{RBSI}}_i={\displaystyle \sum _{k=1}^8}{\omega }_k{g}_{k,i},{\omega }_k\ge 0,{\displaystyle \sum _{k=1}^8}{\omega }_k=1$$

(1)

where $g_{k,i}$ is the normalized score of unit i for component k, and ${\omega }_k$ is the corresponding weight. The eight components are retained capacity ($g_{1,i}$), impedance and internal-resistance behavior ($g_{2,i}$), thermal behavior under load ($g_{3,i}$, inferred from non-destructive proxies), inter-cell homogeneity ($g_{4,i}$), round-trip efficiency ($g_{5,i}$), self-discharge ($g_{6,i}$), traceability ($g_{7,i}$), and safety/risk ($g_{8,i}$). Each component is normalized to $[0,1]$, with higher values indicating more favorable outcomes. Distinct symbols ($\omega$ and g) distinguish this battery-level index, Equation (1), from the pathway-level IVI defined in Equation (15). The weights ${\omega }_k$ are not fixed here: they are to be grounded in the literature, elicited through a structured expert procedure (AHP/BWM/Delphi [34,35,37]), and varied in sensitivity analysis; the index is to be reported together with its weighting rationale and its sensitivity envelope.

The RBSI is a proposed decision-support index, not a universal normative threshold. Composite indexing is an established technique [42] and is not claimed as novel; the contribution lies in the battery-specific eight-component set and in its coupling to the pathway-level IVI. Any triage thresholds applied to the components are research thresholds, not operational cut-offs: an indicative Tier A/Tier B/Exclusion stratification, requiring empirical calibration before use, is tabulated in the Supplementary Materials. The same tables set out the proposed mapping of the eight components onto the six IVI criteria and the role of $g_8$ alongside C2 and $R_i$—a delineation intended to keep the same safety evidence from being counted twice at the battery–pathway interface. The common ~80% SoH retirement convention is noted but rejected as a sole criterion [24]: a single capacity threshold carries none of the safety, homogeneity, or traceability information that the remaining components are designed to capture.

3. Mathematical Model

This section specifies the pathway-level model as Equations (2)–(16); the battery-level RBSI of Section 2.7 carries Equation (1). It couples a failure mode, effects, and criticality analysis (FMECA) risk layer [30,31] with a multi-criteria decision-analysis (MCDA) ranking layer [32,33,34,35], adds a screening-level avoided-burden environmental proxy that derives the environmental-benefit criterion, and combines these into a single integrated viability index (IVI). The pathway-level expressions are introduced here in order as Equations (2)–(16); other sections refer to them by these numbers and introduce no new numbered equations.

All numeric inputs reported below, the performance matrix entries, the FMECA severity, occurrence, and detection ratings, the environmental screening parameters, the scenario weight vectors, the risk-penalty coefficient, and the decision thresholds, are illustrative and transparently assumed or literature-informed, not measured. They are provided so the formal machinery can be exercised end-to-end. No claim of measurement, calibration, or statistical representativeness is made. The framework is designed for application in data-scarce emerging-market contexts, where structured decision methods substitute for the operational data that are typically unavailable [36,43,44,45]. Future empirical and expert-elicitation work should refine these illustrative inputs with context-specific values (Section 2.3 and Section 6).

3.1. Alternatives

The decision set comprises four mutually exclusive end-of-use pathways for a returned electric-vehicle (EV) or plug-in hybrid electric-vehicle (PHEV) battery pack:

$$A=\{A_1,A_2,A_3,A_4\}$$

(2)

where A1 = Refurbishment (restoring the pack for automotive reuse), A2 = Second-life reuse (repurposing a degraded automotive pack into a less-demanding stationary-storage application), A3 = Recycling (material recovery), and A4 = Rejection or safe disposal (the value-destroying fallback retained for completeness). The alternatives are ordered, conceptually, along the circular-economy value-retention hierarchy: refurbishment and second-life reuse preserve embodied value and energy, recycling recovers only materials, and rejection destroys residual value [4,5,6,7]. The model treats the four alternatives symmetrically, each is scored on the same criteria and carries its own FMECA risk penalty, and selects among them rather than presupposing a single preferred pathway.

3.2. Criteria and Indicators

The alternatives are evaluated on six benefit criteria (higher scores preferable on every criterion):

$$C=\{C_1,C_2,C_3,C_4,C_5,C_6\}$$

(3)

namely C1 Technical viability, C2 Inherent/designed safety and reliability, C3 Economic viability, C4 Environmental benefit, C5 Regulatory readiness, and C6 Market/implementation readiness. These operationalize the five families identified in RQ1: residual technical performance and state of health (SoH) [16,19]; inherent/designed safety and reliability [8,30]; economic viability including cost and residual value [9,29]; environmental benefit screened with life-cycle impact-assessment logic [11,12,13,14,15]; regulatory readiness reflecting the EU battery regulation and the Digital Product Passport (DPP) [20,21,22,23]; and market/implementation readiness capturing consumer acceptance and infrastructure [1,16,17,18,28]. The less obvious mappings of the recurring decision dimensions onto this criterion set are worth making explicit: refurbishment feasibility enters through the technical-viability criterion and the threshold screening of Section 3.8, traceability and regulation through C5, and infrastructure through C6.

Operational failure risk is deliberately not modeled as a seventh weighted criterion. Treating risk as one more benefit criterion would conflate the likelihood and consequence of failure with the desirability of an outcome and would allow a high benefit score to mask a serious safety hazard. Instead, residual operational risk is quantified separately through the FMECA layer (Section 3.3) and enters the integrated index as an explicit penalty term (Section 3.7), keeping the benefit and risk dimensions interpretable and consistent with risk-informed practice [30,31,45].

The performance of each alternative on each criterion is recorded in the decision matrix $X=\left[x_{ij}\right]$, rows indexed by alternative i and columns by criterion j. The illustrative entries, on a 1–9 benefit scale, are given in

Table 2. Illustrative performance matrix X (1–9 scale, all benefit criteria). Values are assumed or literature-informed, not measured; C2 is inherent/designed safety (Section Construct Partition: Inherent Safety (C2) Versus Residual Operational Risk ($R_i$)) and the C4 column is computed by the screening of Section 3.6.

Alternative	C1 Tech	C2 Safety	C3 Econ	C4 Env	C5 Reg	C6 Market
A1 Refurbish	9	5	8	6.0	6	6
A2 Second-life	7	6	6	9.0	6	6
A3 Recycle	4	8	4	4.0	9	8
A4 Reject/Disposal	2	6	1	1.4	7	3

. The C4 column is not assumed directly; it is derived from the screening-level avoided-burden proxy of Section 3.6.

The pattern reflects the qualitative logic of the pathways: refurbishment scores highest on technical and economic viability but only moderately on inherent safety, since residual hazard and information asymmetry are most acute when a degraded pack re-enters automotive service [16,19]; recycling scores highest on inherent safety and regulatory readiness as the most institutionally established pathway, conducted in controlled industrial facilities [6,7,20]; and rejection scores low on every value-bearing criterion.

3.3. FMECA Risk Model

The FMECA layer identifies the dominant failure modes of each pathway and converts them into a normalized, comparable risk penalty [30,31]. The inventory operationalizes the pathway-specific hazards documented in the literature: undetected cell degradation and SoH mis-estimation with residual thermal-runaway potential for refurbishment, which re-enters demanding service under the information asymmetry of used-product markets [16,19]; accelerated degradation under a changed duty cycle and battery-management-system (BMS) or system-integration incompatibility for second-life roles [9,10]; process emissions, hazardous-material handling, and end-of-life transport hazard for recycling [6,8]; and loss of embodied value with improper-disposal risk for rejection [6,8]. For each failure mode i, a criticality number is the product of three 1–10 expert-judgment ratings, severity $S_i$, occurrence $O_i$, and detection difficulty $D_i$:

$$C{n}_i=S_i\times O_i\times D_i$$

(4)

Because the maximum attainable criticality is $10\times 10\times 10=1000$, each criticality number is normalized to the unit interval:

$$R_i={\displaystyle \frac{C{n}_i}{1000}}\in [0,1]$$

(5)

The risk penalty of a pathway is the mean of the normalized criticalities of its failure modes. The arithmetic mean is adopted as the primary reporting baseline not because it is superior to alternative aggregations, but because it is the most transparent one available and yields penalties directly comparable across pathways whose failure-mode inventories differ in size (two to four modes here); unlike a maximum-based or blended index, it introduces no additional aggregation parameter that would itself require calibration. Writing ${\mathcal{F}}_k$ for the set of failure modes of pathway $A_k$ and $|{\mathcal{F}}_k|$ for its cardinality:

$$R_{A_k}={\displaystyle \frac{1}{|{\mathcal{F}}_k|}}\sum _{i\in {\mathcal{F}}_k}{R}_i$$

(6)

The illustrative failure modes and ratings are listed in

Table 3. Illustrative FMECA failure modes with severity (S), occurrence (O), detection difficulty (D), criticality ($Cn$, Equation (4)), and normalized criticality (R, Equation (5)). Ratings are assumed on 1–10 scales, not measured.

Pathway	Failure Mode	S	O	D	Cn	R
A1 Refurbish	Undetected cell degradation/SoH mis-estimation	8	5	6	240	0.240
A1 Refurbish	Thermal runaway after refurbishment	9	3	5	135	0.135
A1 Refurbish	Provenance/usage-history information asymmetry	5	6	7	210	0.210
A1 Refurbish	Warranty and liability gap	6	5	4	120	0.120
A2 Second-life	Accelerated degradation under new duty cycle	6	6	5	180	0.180
A2 Second-life	BMS/system-integration incompatibility	6	5	6	180	0.180
A2 Second-life	Residual fire risk in stationary use	8	3	5	120	0.120
A3 Recycle	Process emissions/hazardous-material handling	6	4	3	72	0.072
A3 Recycle	Material-recovery yield variability	4	5	4	80	0.080
A3 Recycle	Transport hazard of end-of-life packs	7	4	5	140	0.140
A4 Reject/Disposal	Embodied resource/value loss	7	8	2	112	0.112
A4 Reject/Disposal	Improper disposal/environmental contamination	8	4	4	128	0.128

, on the conventional 1–10 FMECA scales, not field-measured frequencies.

Applying Equation (6) yields the per-pathway risk penalties used throughout: A1 Refurbish $R_{A_1}=0.176$; A2 Second-life $R_{A_2}=0.160$; A3 Recycle $R_{A_3}=0.097$; A4 Reject $R_{A_4}=0.120$. Under these illustrative inputs, refurbishment carries the highest residual operational risk, driven by undetected degradation and information asymmetry, while recycling carries the lowest, reflecting the maturity of material-recovery processes [6,7].

Because Equation (6) takes the arithmetic mean of a pathway’s normalized criticalities, a single catastrophic failure mode can be diluted by lower-criticality modes. Two safeguards address this. First, a non-compensatory catastrophic-severity gate is applied in parallel with the mean: any pathway containing a failure mode with severity $S\ge 8$ is flagged for mandatory safety review and cannot be cleared on a low aggregated penalty alone, consistent with the absolute gate of Equation (16). Second, a severity-aware blended penalty $R_{A_k}=\alpha {\overline{R}}_{A_k}+(1-\alpha ){\max }_{i\in {\mathcal{F}}_k}{R}_i$ (illustratively $\alpha =0.5$) tests sensitivity to the aggregation rule; its results, reported in Section 4.3, leave every scenario leader unchanged. The inventory itself is deliberately non-exhaustive: candidate modes such as lithium plating within undetected degradation, interconnect and busbar faults introduced at pack reopening, and thermal events at recycling facilities are left to the Delphi refinement round (Section 2.3), and the A4 entries are pathway-inherent burdens treated as failure modes for symmetry. The aggregation choice affects penalty magnitudes and, because the ratings are expert-judgment-based rather than incident-derived, requires calibration in future work. For the same reason, the classical $S\times O\times D$ formulation is retained over fuzzy FMECA for this conceptual demonstration: the study does not yet possess the calibrated expert judgments or empirically grounded membership functions on which a defensible fuzzy model would rest, whereas the crisp arithmetic can be audited step by step. Fuzzy extensions are better suited to representing linguistic uncertainty in expert ratings [31], and their evaluation is scheduled alongside the Delphi elicitation of Section 2.3.

Construct Partition: Inherent Safety (C2) Versus Residual Operational Risk ($R_i$)

A central construct-validity concern for any framework that scores safety as a benefit while penalizing risk is the danger of counting safety twice. The model partitions the safety construct along two distinct axes. The benefit criterion C2 captures the inherent/designed safety and reliability of a pathway, its intrinsic safety architecture and process maturity. Recycling, for example, occurs in controlled industrial facilities with established containment, whereas in-field refurbishment depends on the quality of available diagnostics; C2 therefore rewards the structural safety advantage a pathway possesses by design, before any specific failure mode is realized. The FMECA penalty $R_i$ (Equation (6)), by contrast, captures the residual operational failure exposure that remains after a pathway’s designed controls are applied, the aggregate criticality of failure modes that can still be realized in operation.

C2 (designed safety) and $R_i$ (residual realized risk) are thus distinct axes: one measures how safe a pathway is built to be, the other how much failure exposure survives that design. Routing residual risk exclusively through $R_i$, rather than embedding it as a seventh weighted criterion, keeps the two axes separate by construction; some residual overlap is unavoidable, because certain failure modes (for example, thermal runaway) have an evident safety character. Whether the conclusions depend on that overlap is tested by recomputing the framework with the C2 benefit removed (Section 4.3).

3.4. MCDA Normalization and Weighted Sum

The MCDA layer first normalizes X: each benefit-criterion entry is divided by the column maximum,

$$r_{ij}={\displaystyle \frac{x_{ij}}{{\max }_i{x}_{ij}}}$$

(7)

with the min-based cost form stated for completeness, all six criteria here being benefit-type:

$$r_{ij}={\displaystyle \frac{{\min }_i{x}_{ij}}{x_{ij}}}$$

(8)

The aggregate benefit score under the weighted-sum model is:

$$V_i={\displaystyle \sum _{j=1}^n}{w}_j{r}_{ij}$$

(9)

where the weight vector $w=(w_1,\dots ,w_n)$ encodes criterion importance, constrained to sum to unity, with equal weighting as the baseline:

$${\displaystyle \sum _{j=1}^n}{w}_j=1,w_j={\displaystyle \frac{1}{n}}\left(\mathrm{baseline}\right)$$

(10)

The equal-weight baseline is used as a neutral benchmark when no validated stakeholder-preference vector is available, avoiding an unsupported normative priority in the illustrative setup. The weights are scenario-based rather than empirically derived; the scenario vectors of Section 3.9 represent distinct, internally consistent stakeholder priorities, and AHP [34], BWM [35], and Delphi weighting [37] are the future routes to context-specific values.

3.5. TOPSIS Ranking

To test whether the rankings are an artifact of a single aggregation rule, the model also implements the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) [32], which uses vector normalization rather than Equation (7):

$$v_{ij}={\displaystyle \frac{x_{ij}}{\sqrt{{\sum }_i{x}_{ij}^2}}}$$

(11)

The normalized values are weighted, ${\widehat{v}}_{ij}=w_j{v}_{ij}$, and the positive-ideal ${\widehat{v}}_j^{+}={\max }_i{\widehat{v}}_{ij}$ and negative-ideal ${\widehat{v}}_j^{-}={\min }_i{\widehat{v}}_{ij}$ solutions are identified for each benefit criterion. The Euclidean distances of each alternative to the two ideal solutions are:

$$D_i^{+}=\sqrt{\sum _j{\left({\widehat{v}}_{ij}-{\widehat{v}}_j^{+}\right)}^2},D_i^{-}=\sqrt{\sum _j{\left({\widehat{v}}_{ij}-{\widehat{v}}_j^{-}\right)}^2}$$

(12)

The closeness coefficient is:

$$C{C}_i={\displaystyle \frac{D_i^{-}}{D_i^{+}+D_i^{-}}}$$

(13)

Alternatives are ranked by descending $C{C}_i$. TOPSIS is computed across all six scenarios of Section 3.9, and because it carries no explicit risk penalty, its agreement and its divergence with the IVI are equally informative; the comparison is analyzed in Section 4.1.

3.6. Screening-Level Avoided-Burden Proxy for the Environmental Criterion (C4)

The environmental-benefit criterion C4 is not assigned as a free score; it is derived from a transparent, literature-anchored screening-level proxy for the avoided manufacturing burden of each pathway. The proxy estimates the cradle-to-gate greenhouse-gas (GHG) burden of a new pack that each pathway notionally displaces, then maps it onto the 1–9 scale of C4. For an illustrative 60 kWh pack using NMC811 chemistry, with a manufacturing carbon intensity of 74 kg CO₂e/kWh (the P50 midpoint reported for NMC811) [12], the embodied cradle-to-gate footprint is approximately 4440 kg CO₂e. The avoided-burden proxy for pathway i is:

$$EN{V}_i=d_i\times \mathrm{pack}\text{\_}\mathrm{kWh}\times \mathrm{carbon}\text{\_}\mathrm{intensity}$$

(14)

where $d_i$ is a literature-informed displacement fraction, the share of new-pack manufacturing burden notionally avoided by pathway i [9,10].

What the proxy is and is not. Equation (14) compares the four pathways on one indicator only: cradle-to-gate manufacturing CO₂e notionally avoided for a single illustrative pack, relative to a common new-pack counterfactual. It is not a life-cycle assessment. It adopts a per-pack basis rather than a service-based functional unit (such as kg CO₂e per kWh delivered in second-life service); it omits the use phase and the collection, transport, diagnostics, refurbishing, and end-of-life processing stages; it relies on a single point estimate of manufacturing carbon intensity rather than a regionalized inventory; and it credits recycling only through a displacement fraction rather than a characterized recovery model. The proxy therefore cannot establish the environmental superiority of any pathway; it supplies only an illustrative score for criterion C4. Any environmental claim would require a full ISO 14040/14044-compliant assessment with a service-based functional unit, complete inventory, and characterization (for example, ReCiPe 2016 [15]), specified as future work (Section 2.3). The proxy inputs, avoided CO₂e, and resulting C4 scores are given in

Table 4. Screening-level avoided-burden proxy driving criterion C4 (Equation (14)). Illustrative pack: 60 kWh NMC811; manufacturing intensity 74 kg CO₂e/kWh (P50) [12]; embodied ≈4440 kg CO₂e. Displacement fractions are literature-informed midpoints [9,10]. Screening-level proxy, not a full ISO 14040/14044 life-cycle assessment.

Pathway	Displacement Fraction	Avoided kg CO2e	C4 (1–9)
A1 Refurbish	0.25	1110	6.0
A2 Second-life	0.40	1776	9.0
A3 Recycle	0.15	666	4.0
A4 Reject/Disposal	0.02	89	1.4

.

The fractions order the pathways plausibly: refurbishment typically replaces only degraded modules [4,11] and credits a shorter displacement horizon than a full new pack, so its fraction sits below that of second-life reuse, while recycling is credited for material recovery only. The avoided-CO₂e column is mapped onto the 1–9 scale by an affine transformation (C4 ≈ 1.0 + 0.0045 × avoided kg CO₂e), a scaling convenience without physical meaning. Under min–max and logarithmic monotone alternatives to this mapping, the C4 rank order (A2 > A1 > A3 > A4) and the five priority-scenario leaders are unchanged; only the baseline near-tie between the two value-retention pathways can flip, consistent with the perturbation analysis of Section 4.2.

3.7. Integrated Viability Index (IVI)

The central artifact couples the MCDA benefit layer with the FMECA risk layer into a single scalar per alternative. For alternative i, let $C_{ij}$ denote its normalized benefit-criterion score on criterion j (Equation (7)) and $R_i$ its normalized FMECA risk penalty (Equation (6)). The integrated viability index is:

$$IV{I}_i={\displaystyle \sum _{j=1}^n}{w}_j{C}_{ij}-\lambda R_i$$

(15)

The first term is the weighted benefit score; the second subtracts residual operational risk, scaled by the risk-penalty coefficient $\lambda \ge 0$. The coefficient acts as an inverse proxy for the effectiveness of risk-mitigation mechanisms, namely warranty coverage, independent SoH certification, and the DPP [20,21,22,23]: where such mechanisms credibly contain residual risk the effective $\lambda$ is low and the benefit term dominates; where they are weak, risk-laden pathways are penalized heavily. The baseline $\lambda =0.30$ is adopted as a mid-range value between the two preference-transition bands identified in Section 4.3, representing partially credible mitigation; it is a tunable penalty weight rather than a universal constant, so the specific transition values are functions of the illustrative inputs, and baseline results are read jointly with the full $\lambda$ sweep. Because $R_i$ alone carries residual operational risk, the IVI does not re-penalize the inherent safety already scored in C2 (Section Construct Partition: Inherent Safety (C2) Versus Residual Operational Risk ($R_i$)).

Algebraically, Equation (15) with fixed $\lambda$ is equivalent to a weighted-sum model in which risk enters as an additional criterion with negative weight; the separation is architectural rather than mathematical. Its value is interpretive: the benefit weights are normalized over benefits alone, so risk does not compete for weight mass; the penalty strength is swept explicitly (Section 4.3) rather than embedded in a weight vector; and the non-compensatory constraint that no benefit can rescue a safety-critical pack is enforced by the gate of Equation (16), not by the penalty term.

3.8. Threshold Decision Model

Whereas the IVI ranks alternatives on a continuous scale, the framework also provides a complementary rule-based screening logic mapping directly observable pack attributes, state of health, aggregate risk, and safety score, onto a recommended pathway, applied in lexicographic order:

$$\mathrm{decision}=\left\{\begin{array}{cc}\mathrm{Reject}/\mathrm{safe}\mathrm{disposal}\hfill & \mathrm{if}\mathrm{risk}>0.70\mathrm{or}\mathrm{safety}\mathrm{critically}\mathrm{low}\hfill \\ \mathrm{Refurbish}\hfill & \mathrm{else}\mathrm{if}\mathrm{SoH}\ge {\theta }_{\mathrm{ref}}\mathrm{and}{R}_i\le R_{\max }\hfill \\ \mathrm{Second}\text{-}\mathrm{life}\mathrm{reuse}\hfill & \mathrm{else}\mathrm{if}{\theta }_{\mathrm{second}}\le \mathrm{SoH}<{\theta }_{\mathrm{ref}}\mathrm{and}\mathrm{safety}\ge {\theta }_{\mathrm{safety}}\hfill \\ \mathrm{Recycle}\hfill & \mathrm{otherwise}\hfill \end{array}\right.$$

(16)

The illustrative threshold parameters are ${\theta }_{\mathrm{ref}}=0.80$, ${\theta }_{\mathrm{second}}=0.60$, ${\theta }_{\mathrm{safety}}=0.60$, and $R_{\max }=0.40$; here risk denotes a pack-level aggregate risk score on $[0,1]$ produced by the screening assessment, distinct from the pathway penalty $R_i$. That ${\theta }_{\mathrm{ref}}$ coincides with the conventional ~80% retirement value is deliberate and does not reinstate the single-criterion convention rejected in Section 1: the threshold operates only in conjunction with the risk and safety gates and the full multi-criteria evaluation, never as a stand-alone pass/fail rule. The logic encodes the value-retention hierarchy: a high-SoH, low-risk pack is steered to refurbishment, a moderately degraded but still-safe pack to second-life reuse, and the remainder to recycling, with the absolute gate diverting hazardous packs to safe disposal regardless of residual value. The screening and the IVI are mutually reinforcing, thresholds providing fast triage while the IVI resolves finer trade-offs; the gates are exercised on concrete SoH inputs in Section 4.3, and the values are to be calibrated against field data.

3.9. Scenario and Sensitivity-Analysis Design

Because the criterion weights are not empirically fixed, the framework is exercised across a set of scenarios, each a distinct but internally coherent weighting of stakeholder priorities. Six scenarios are defined: a baseline of equal weights (S0, Equation (10)) and five priority scenarios that each elevate one criterion family while the weights still sum to unity (Equation (10)). The vectors are given in

Table 5. Scenario weight vectors (${\sum }_j{w}_j=1$). Weights are illustrative representations of stakeholder priorities, constructed to span the criterion families, not empirically elicited.

Scenario	C1	C2	C3	C4	C5	C6
S0 Baseline (equal)	0.167	0.167	0.167	0.167	0.167	0.167
S1 Safety-priority	0.12	0.34	0.12	0.12	0.18	0.12
S2 Cost-priority	0.16	0.14	0.34	0.12	0.12	0.12
S3 Environmental-priority	0.12	0.14	0.12	0.34	0.16	0.12
S4 Regulatory-strict	0.10	0.20	0.10	0.14	0.34	0.12
S5 Infrastructure-constrained	0.14	0.16	0.16	0.10	0.10	0.34

. These scenarios were deliberately constructed to span the criterion families, so the finding that no single pathway dominates is demonstrated by design as much as it is emergent; this is stated explicitly to avoid overclaiming the result.

For each scenario, the IVI of every alternative is computed via Equation (15) with the scenario-specific weights and the baseline $\lambda =0.30$, directly addressing RQ3; the rankings are reported in Section 4.

Four complementary checks probe the rankings: a risk-penalty sensitivity that varies $\lambda$ under baseline weights; a one-at-a-time weight sensitivity that perturbs each criterion weight in turn (re-normalizing the remainder to preserve Equation (10)); a safety double-counting check that recomputes the IVI with the C2 benefit removed, so safety enters only through $R_i$ (Section Construct Partition: Inherent Safety (C2) Versus Residual Operational Risk ($R_i$)); and an input-score robustness check of 20,000 Monte-Carlo perturbations of the performance matrix, each score independently jittered by a uniform $U[-1,+1]$ draw and clipped to $[1,9]$, under the declared random seed (see the Data Availability Statement). Because criterion C4 is derived through Equation (14), jittering it directly severs, for this check only, its link to the displacement fractions; the perturbation is read as generic scoring noise on all six criteria. Results are reported in Section 4.2 and Section 4.3. Together these analyses map the decision space rather than issuing a single deterministic recommendation. Named methodological extensions (fuzzy FMECA [31], VIKOR [33], and real-options timing [43,44]) are catalogued in the Supplementary Materials.

4. Illustrative Results

This section exercises the full machinery, the IVI, TOPSIS, the FMECA penalty, and the scenario and sensitivity checks, on a single illustrative pack. All inputs are those of Section 3: the performance matrix (Table 2) with C4 derived through Equation (14), the FMECA penalties of Equation (6) on conventional 1–10 scales consistent with practice discussed in [30,31], the six weight vectors of Table 5, and the baseline $\lambda =0.30$. The results demonstrate the decision logic under stated assumptions and are not recommendations for any real battery population; Figure 1, Figure 2, Figure 3 and Figure 4 report them.

4.1. Pathway Profiles and Conditional Preference

The scores of Table 2, Table 3 and Table 4 give each pathway a distinct profile, and the IVI converts those profiles into conditional preference. Refurbishment offers the largest value-retention upside but concentrates operational and trust-related risk: its penalty, the highest of the four (Figure 3), is driven by undetected degradation and provenance asymmetry, both with high detection difficulty, consistent with the literature on information asymmetry in refurbished markets [4,5,11,16,19,28]. It leads under cost-priority weighting, and its preferred status is contingent on economic or technical emphasis and on credible containment of the penalized risks. Second-life reuse carries the strongest proxy-derived environmental score (Table 4) and moderate scores elsewhere; its penalty reflects the engineering uncertainty of repurposing (duty-cycle change, battery-management-system integration) rather than acute information asymmetry [9,10]. It leads under environmental priority and is otherwise the mid-range compromise. Recycling combines the strongest designed-safety, regulatory, and market scores with the lowest penalty, consistent with its status as the most established and most explicitly regulated route [6,7,8,20]; it becomes preferred under safety-priority, regulatory-strict, and infrastructure-constrained weightings. Rejection is never preferred in any scenario, its IVI remaining lowest throughout, in line with value-retention logic [1,4,5]. Scenario-level IVI values are consolidated in Section 4.2 and Figure 1.

The TOPSIS comparison is informative precisely because the two rules do not coincide. They agree on the leading pathway in three of six scenarios (baseline and environmental priority, led by second-life reuse; cost priority, led by refurbishment) and agree that rejection is last in all six ($C{C}_4\le 0.164$). They diverge in the safety-priority, regulatory-strict, and infrastructure-constrained scenarios: the IVI ranks recycling first, whereas TOPSIS, which subtracts no risk penalty, still favors a value-retention pathway, and recycling never attains the top TOPSIS closeness in any scenario. A controlled ablation on the same matrix and weights isolates three mechanisms. First, adding the explicit penalty $\lambda R_i$ changes the leader in S5 (from A1 to A3) under both column-maximum and vector-normalized weighted sums; in S1 and S4 it preserves A3 as leader. Second, switching weighted-sum normalization (column-maximum to vector) changes magnitudes but not the leader in S1, S4, and S5. Third, TOPSIS still selects A2 in S1 and S4 despite the same weighted, vector-normalized inputs, indicating that the distance-to-ideal aggregation rule itself, not only normalization, contributes to divergence. Closeness values are also sensitive to the composition of the alternative set, a known property of distance-to-ideal methods [32,33]. The full closeness coefficients are tabulated in

Table 6. Condensed TOPSIS closeness coefficients (all scenarios; reproduced from the Supplementary reproducibility script).

Scenario	A1	A2	A3	A4
S0 Baseline (equal)	0.69	0.70	0.48	0.08
S1 Safety-priority	0.55	0.61	0.57	0.16
S2 Cost-priority	0.81	0.71	0.44	0.05
S3 Environmental-priority	0.63	0.81	0.41	0.06
S4 Regulatory-strict	0.53	0.58	0.57	0.16
S5 Infrastructure-constrained	0.68	0.65	0.63	0.07

; Figure 4 displays them by scenario.

4.2. Comparative Ranking Under Alternative Assumptions

The substantive result is not that different pathways lead in different scenarios, which the scenario construction makes near-inevitable, but that the framework localizes where, why, and at what risk penalty each change of preference occurs. Under the stated assumptions the leading pathway varies as follows (Figure 1): at baseline S0 the two value-retention pathways jointly lead, second-life reuse marginally ahead of refurbishment (0.734 versus 0.732), both above recycling (0.702) and disposal (0.365); refurbishment leads under cost-priority S2 (0.785); second-life reuse under environmental-priority S3 (0.777); and recycling under safety-priority S1 (0.777), regulatory-strict S4 (0.787), and infrastructure-constrained S5 (0.757). Rejection is preferred in none, its IVI ranging from 0.304 to 0.480 and remaining lowest in every case.

The baseline ordering between the two value-retention pathways lies within scoring noise and must not be over-read. Across the 20,000 perturbed replicates of Section 3.9, second-life reuse leads in 50.9% of runs, refurbishment in 43.1%, recycling in 5.9%, and rejection in none, conditional on the stated perturbation model, with a median top-two margin of about 0.022. The substantive baseline finding is that the two value-retention pathways jointly dominate recycling and disposal; which of the two leads is inside the noise.

A one-at-a-time examination of the weights clarifies the broader pattern. From the equal-weight baseline, boosting C2, C5, or C6 moves the lead toward recycling; boosting C1 or C3 moves it toward refurbishment; and boosting C4 moves it toward second-life reuse. These shifts are intuitive given the performance profiles (recycling carries the strongest safety, regulatory, and market scores with the lowest penalty, while second-life carries the strongest environmental score) and indicate that the ranking responds coherently to the criterion emphasized rather than erratically.

The framework does not produce, and is not intended to produce, a context-free best pathway; the orderings characterize the behavior of the decision logic under illustrative inputs, not the relative merit of real battery pathways.

4.3. Sensitivity and Robustness Analysis

The checks specified in Section 3.9 are reported here in turn.

The $\lambda$ sensitivity (Figure 2) traces each pathway’s IVI under baseline weights as $\lambda$ increases from zero. Because refurbishment carries the highest penalty and recycling the lowest (Table 3), leadership migrates: from refurbishment to second-life reuse between $\lambda \approx 0.10$ and $0.15$, and from second-life reuse to recycling between $\lambda \approx 0.80$ and $0.85$. Since $\lambda$ proxies mitigation credibility (Section 3.7), the transitions state the conditions under which refurbishment is preferred, not an actionable magnitude of risk containment.

Removing the C2 benefit entirely, so that safety enters only through $R_i$, preserves the non-dominance structure, with three productive pathways still leading across scenarios and rejection never leading, but changes the leader in two of six scenarios (safety-priority and infrastructure-constrained, where refurbishment overtakes recycling). The conditionality is therefore not an artifact of counting safety twice, while the scenario-level leaders are C2-sensitive: C2 carries decision-relevant information about designed safety beyond what the residual-risk penalty captures (Section Construct Partition: Inherent Safety (C2) Versus Residual Operational Risk ($R_i$)).

Recomputing the penalty with the severity-aware blended rule of Section 3.3 gives $R_{A_1}=0.208$, $R_{A_2}=0.170$, $R_{A_3}=0.119$, and $R_{A_4}=0.124$, and re-running the IVI leaves the leading pathway in every scenario unchanged: only the magnitude of the penalty, not the ordering, depends on the aggregation choice. An SoH sweep of the threshold rules (Equation (16)) at a fixed safety level of 0.65 and aggregate risk of 0.30 fires the gates as specified: SoH ≥ 0.80 routes to refurbishment, 0.60 ≤ SoH < 0.80 to second-life reuse, SoH = 0.55 (below ${\theta }_{second}$) to recycling, and safety-critical risk above 0.70 to rejection.

Table 7. Condensed threshold-sweep outcomes under Equation (16) (illustrative: fixed safety = 0.65, aggregate risk = 0.30 unless noted).

SoH	Safety	Aggregate Risk	Assigned Pathway
0.90	0.65	0.30	Refurbish
0.80	0.65	0.30	Refurbish
0.70	0.65	0.30	Second-life reuse
0.60	0.65	0.30	Second-life reuse
0.55	0.65	0.30	Recycle
0.70	0.65	0.75	Reject/safe disposal

provides a compact subset of the sweep in the main text.

The checks converge. Pathway preference remains conditional; the two value-retention pathways jointly lead at baseline, their internal order sitting inside scoring noise; leadership migrates with $\lambda$; rejection never leads; and the structure survives both C2 removal and the alternative FMECA aggregation. These remain demonstrations under assumed inputs, not findings about real batteries.

5. Discussion

The illustrative results of Section 4 demonstrate how the framework behaves; they are not empirical findings about any battery fleet. Two structural features carry the interpretive weight: the divergence between the risk-penalized IVI and penalty-free TOPSIS is localized in exactly the scenarios where the risk treatment decides, and preference migrates traceably with the risk-penalty coefficient. The conditional, no-dominance pattern frames both.

5.1. Delimitation of the Contribution (Existing/Adapted/Original)

To forestall overclaiming, the contribution is delimited in three categories. Existing (cited, used as-is): multimodal electrochemical and thermal diagnostics [25,26,27]; active reconditioning [47]; degradation and remaining-useful-life prediction [46,62,63]; validation statistics [58,59,60]; the life-cycle-assessment framework [40,41]; techno-economic practice [61]; open battery-data formats [57]; the relevant repurposing, safety, EV-test, and transport standards [20,49,50,51,52,53,54,55,56]; and the decision-layer methods themselves—FMECA [30,31], the MCDA family including TOPSIS, VIKOR, AHP, and BWM [32,33,34,35], life-cycle impact assessment (LCIA) characterization [15], Delphi elicitation [37], and composite indexing [42]. Adapted (cited, with the adaptation stated): the assembly of these elements into a screening → diagnostic → refurbish → prognosis → assessment architecture targeted specifically at refurbished (not only second-life) batteries; the triage thresholds; and the emission-factor-to-criterion mapping. Original (proposed): the RBSI eight-component eligibility index and its coupling to the pathway-level IVI; the active-versus-passive refurbishing comparative design; and the uncertainty propagation linking diagnostics to the techno-economic, life-cycle, and risk decision. No single component is claimed as a new individual method; the novelty resides in the RBSI–IVI coupling.

A fourth boundary separates what is specified from what has been carried out: the Delphi calibration, the execution of the diagnostic sublayer, and the ISO 14040/14044-compliant life-cycle assessment are future work (Section 2.3 and Section 6), and no result reported here rests on them.

5.2. Theoretical Implications

The central contribution is the formalization of conditional viability: whether to refurbish, reuse in second-life, recycle, or reject has no context-free answer but depends on how the criteria are weighted and how operational risk is penalized (Section 4.2, Figure 1). This reframes the circular-economy value-retention hierarchy [4,5,6,7] not as a fixed ranking but as a conditional ordering whose realization depends on context, retaining the logic of preserving embodied value before material recovery [4,9,11]. Read at the system level, pathway selection is an emergent property of how a socio-technical system weights its technical, safety, economic, environmental, regulatory, and market signals and how it penalizes residual risk, rather than a fixed attribute of the pack; the framework makes those dependencies explicit and traceable, which is the level at which governance, certification, and infrastructure interventions actually operate.

A second implication separates risk from benefit: routing operational failure risk through a distinct FMECA penalty rather than a weighted criterion distinguishes how good a pathway is when it works from how likely and consequential failure is, and locates conditional preference in the credibility of risk-mitigation mechanisms rather than in intrinsic pathway merit. As the $\lambda$ sweep shows (Section 4.3), refurbishment and second-life reuse are the value-retaining choices only where the residual risks of undetected degradation, thermal events, provenance asymmetry, and warranty gaps are credibly contained. That rejection is never preferred reproduces the intuition that value-destroying disposal is a fallback of last resort [4], positioning the contribution within work on circular automotive decision contexts and lifecycle management of used EV batteries [1,38] and on residual value and information asymmetry in refurbished markets [16,19,28,29].

5.3. Methodological Implications

The framework integrates structured decision methods into a single transparent artifact: FMECA supplies a normalized pathway-level risk signal [30,31], weighted-sum and TOPSIS rankings order the alternatives [32,33], and the screening proxy of Equation (14) derives the environmental criterion [15,40,41]. The IVI (Equation (15)) is the integrating device, reconciling benefit and risk through a single tunable coefficient. The double-counting objection, that safety enters once through C2 and again through the safety-related FMECA modes, is addressed by the construct partition (Section Construct Partition: Inherent Safety (C2) Versus Residual Operational Risk ($R_i$)) and its removal check (Section 4.3).

The IVI–TOPSIS relationship is a methodological contrast rather than mutual confirmation: agreement on the robust qualitative findings (rejection never competitive; value-retention pathways dominant under productive weightings) indicates those findings are not an artifact of a single aggregation rule, while the localized divergence isolates the genuine effect of explicitly penalizing residual risk (Section 4.1). The near-tie disclosure of Section 4.2 matters for honest reporting: the baseline ordering between the two value-retention pathways sits inside scoring noise, and stating so guards against over-reading. Methodological refinements remain catalogued for future work (Section 3.9), consistent with the design-science tradition [36].

5.4. Practical Decision-Support Implications

For practitioners, including remanufacturers, fleet operators, recyclers, and second-life integrators, the framework offers a structured way to reason under uncertainty rather than a verdict. The $\lambda$ coefficient is the practical lever, proxying the effectiveness of warranty provision, independent SoH certification, and DPP-based traceability [21,22,23]: where these instruments are credible the effective penalty is low and refurbishment or second-life reuse becomes preferred; where they are weak the penalty rises and the framework steers toward recycling. Investing in diagnostics, certification, and traceability is therefore what makes refurbishment defensible.

A concise implementation workflow is: intake and chain-of-custody registration → safety/traceability gate → multimodal diagnostics and RBSI scoring → stakeholder weight and $\lambda$ selection → IVI/TOPSIS computation → documented pathway assignment and review trigger if gates are violated. Figure 5 renders this workflow as a compact schematic; applied to the illustrative case of Section 4, it is a demonstration of the decision logic, not an industrial case study.

The threshold rules (Equation (16)) encode hard constraints an aggregate score should not override, routing packs with critically low safety or excessive risk directly to rejection. The framework is designed for data-scarce, emerging-market contexts, where measured degradation data, certified provenance, and mature reverse logistics are often unavailable [1,5,8]; by accepting literature-informed inputs and making their consequences visible through sensitivity analysis, it supports a transparent decision that can be revisited as data improve. The infrastructure-constrained scenario, which favors recycling, illustrates how limited refurbishment capacity can dominate otherwise favorable value-retention economics.

5.5. Policy Implications

The policy reading interprets the framework’s behavior and is not an empirical policy evaluation of any specific market. It centers on mechanisms that lower the effective risk penalty and thereby enlarge the conditions under which value-retaining pathways are preferred. The carbon-declaration and battery-passport obligations of Regulation (EU) 2023/1542 (Section 1) institutionalize exactly the transparency instruments the framework treats as risk-mitigation enablers [20,21,23], and blockchain-based refurbishment certification is a complementary credibility instrument [22]; under the stated inputs, effective passport implementation lowers $\lambda$ and shifts the conditional balance toward refurbishment and second-life reuse.

The framework also makes legible the enabler-readiness tension of Section 2.6: the regulatory-strict scenario favors recycling, suggesting that stringent compliance requirements, absent the trust and infrastructure to support refurbishment, can push decisions toward material recovery even where the framework would otherwise rank value retention higher [4,11]. Traceability mandates may therefore deliver their value-retention benefit only when paired with warranty regimes, certification capacity, and reverse-logistics infrastructure; policy attention to that coordination, rather than to either side in isolation, may realize the conditional preference for higher value retention. These statements are hypotheses for empirical policy research, not established prescriptions.

6. Limitations and Future Validation

The framework is conceptual, and its limitations follow from that positioning. No primary empirical results are reported: no expert panel, survey, or battery testing was conducted, and every numeric input (performance scores, FMECA ratings, scenario weights, thresholds, and the coefficient $\lambda$) is illustrative or literature-informed. The scenario results demonstrate the behavior of the decision logic, not the relative merit of real pathways; different but defensible inputs would shift the numbers, and the contribution rests in the reproducible mapping from inputs to a risk-penalized ranking (Equation (15)).

Five structural caveats follow. The baseline ordering of the two value-retention pathways is a statistical near-tie (Section 4.2), so the robust baseline finding is their joint dominance over recycling and disposal, not the ordering between them. The six weighting scenarios were constructed to span the criterion families, so the absence of a dominant pathway is partly by design; what carries evidential weight is the traceable structure of the conditionality, namely which criterion moves the lead and at what risk penalty. The environmental criterion derives from a screening-level avoided-burden proxy (per-pack basis, cradle-to-gate only, single point intensity) and cannot support environmental claims; a full ISO 14040/14044 assessment [15,40,41] with a service-based functional unit is future work. The NMC811-based static displacement fractions used for C4 are not automatically transferable to LFP or other chemistries: manufacturing-intensity baselines, degradation trajectories and service life, recoverable-material value, refurbishing process burdens, and reverse-logistics profiles can differ materially by chemistry and region, and directional ranking changes should not be asserted without chemistry-specific and regionalized inventories. The FMECA layer uses the classical S×O×D structure with assumed ratings, a deliberately non-exhaustive failure-mode inventory, and thresholds that are illustrative parameters rather than normative limits, their appropriate values depending on chemistry, application, regulatory regime, and risk appetite. The severity-aware aggregation and the catastrophic-severity gate leave the scenario leaders unchanged, but penalty magnitudes, and the $\lambda$ values at which preference shifts, depend on ratings yet to be elicited. Finally, some overlap between designed safety (C2) and residual risk ($R_i$) is unavoidable: removing C2 changes the leader in two of six scenarios while preserving the non-dominance structure (Section 4.3). TOPSIS closeness values likewise remain sensitive to the alternative set and normalization [32,33], so the IVI–TOPSIS divergence is reported as a methodological contrast that bounds any single-method claim.

Three scope limitations bound generalizability. The scoping synthesis is structured but not PRISMA-systematic, so omission of adjacent frameworks cannot be excluded. The proposed RBSI and diagnostic sublayer remain unexecuted specifications with uncalibrated weights; first-life history is often unavailable, degrading traceability, and retired-pack populations are heterogeneous, so models calibrated on one cohort may not transfer. Assembling standardized diagnostic data across manufacturers adds further obstacles: SoH definitions and retirement conventions vary [24], access to BMS data and their formats remains fragmented—the motivation for open formats such as the Battery Data Format [57]—and diagnostics must resolve condition at cell, module, and pack level across chemistries and form factors before cohorts become comparable. The uncertainty architecture that remains to be executed is multi-source: measurement uncertainty (sensor/test error), between-cell heterogeneity (true chemical dispersion), expert-judgment uncertainty (weights and S/O/D ratings), and model/scenario uncertainty (aggregation choice, normalization, and pathway assumptions) should be kept separate and propagated hierarchically from cell to module to pack in future validation. Under this proposed protocol, the RBSI would be reported as a distribution or interval rather than a point value, summarized by a conservative statistic (for example, a lower quantile) and guarded by a non-compensatory gate for safety-critical cells or modules, so that chemical outliers—heterogeneous lithium-inventory loss, active-material isolation, or impedance growth concentrated in a few cells—are not masked by averaging; the resulting distributions would then be propagated through the IVI to the techno-economic and life-cycle outputs by Monte Carlo or another justified method, reporting intervals and threshold-crossing probabilities rather than single numbers. None of this propagation has been executed here. The framework targets data-scarce, emerging-market settings, where its transparency is a strength but also a boundary: in data-rich settings, richer probabilistic or option-based methods [43,44,45] may be preferable.

Operationalization is a defined program rather than an aspiration, and its minimum data requirements can be stated compactly: cell-, module-, and pack-level diagnostics with first-life history; real cost data and operating profiles; incident records or expert ratings for the FMECA layer; regionalized, chemistry-specific life-cycle inventories with a service-based functional unit; reverse-logistics, recovery-yield, regulatory-traceability, and market-acceptance data; and external cohorts for validation. The logical sequence is: Delphi-based weight and rating elicitation with AHP/BWM, including refinement of the failure-mode inventory across chemistries (Section 2.3) [34,35,37]; execution of the four-layer diagnostic sublayer with RBSI calibration and validation on external cohorts (Section 2.7); a primary consumer-acceptance survey to replace the literature-informed market-readiness assumptions [16,17,18,19,28]; an ISO 14040/14044-compliant life-cycle assessment with regionalized inventory [11,12,13,14,15,40,41]; techno-economic assessment [61]; a cross-country comparison and an industrial pilot; and ingestion of Digital Battery Passport and certification records [20,21,22,23], with the full protocols in the Supplementary Materials. Until that program is carried out, the framework is a decision-support instrument, not a substitute for physical testing.

7. Conclusions

This paper addressed a decisional question that the battery-circularity literature treats mostly in fragments: under which conditions should a retired EV/PHEV pack be refurbished, redeployed in second-life storage, recycled, or rejected? The answer developed here is architectural. An Integrated Viability Index combines six stakeholder-weighted benefit criteria with a separately computed FMECA residual-risk penalty, scaled by a single coefficient $\lambda$ that makes the role of risk-mitigation credibility explicit; a screening-level avoided-burden indicator positions the environmental criterion without claiming a life-cycle assessment; and TOPSIS provides a methodological contrast that localizes where the explicit risk penalty changes the decision. The five research questions are answered in turn: the six-criterion set (RQ1); the FMECA characterization of pathway-specific failure modes (RQ2); the weighted-sum and TOPSIS rankings with their localized divergence (RQ3); the screening-level environmental embedding (RQ4); and the calibration and measurement architecture of Section 2.3 and Section 2.7 (RQ5).

Exercised on illustrative, literature-informed inputs, the framework yields three structural conclusions. Preference is conditional: refurbishment leads under economic and technical priority with credible risk containment, second-life reuse under environmental priority, and recycling under safety, regulatory, and infrastructure constraints. Preference migrates with risk penalization: as $\lambda$ rises, leadership passes from refurbishment through second-life reuse to recycling, so defensible value retention depends on investment in warranty, certification, and traceability instruments. And rejection never leads, consistent with disposal as a fallback rather than a strategy. These are properties of the decision logic under stated assumptions, not empirical findings about battery fleets.

The proposed RBSI–IVI coupling defines the route from measurement to decision, and its calibration, together with Delphi-based weighting and an ISO-compliant life-cycle assessment, constitutes the validation agenda (Section 6). The framework’s interim value is precisely what data-scarce settings lack: a transparent, reproducible way to reason about circular pathway selection, and to document that reasoning, while the evidence base matures. It supports and structures decisions; it does not replace the physical testing and safety certification on which any deployment must ultimately rest.